ASH1L Links Histone H3 Lysine 36 Dimethylation to MLL Leukemia

Modulation of chromatin structure through covalent histone modifications serves a major role in normal and pathologic gene expression. A significant subset of modifications involves lysine methylation, which is a dynamic and reversible process regulated by two classes of counteracting enzymes: histone lysine methyltransferases and demethylases (1). Methyl groups on histone lysine residues serve as docking sites for reader proteins that bind specific modifications and translate the histone code into various transcriptional outputs (2). Among the diverse family of histone methyltransferases, trithorax (TRX) and absent, small, or homeotic discs 1 (ASH1) are distinguished as trithorax group (trxG) proteins that positively regulate gene transcription and counteract Polycomb group (PcG)–mediated silencing as demonstrated by seminal genetic studies of Drosophila development (3). Furthermore, their mammalian homologs mixed-lineage leukemia (MLL) and ASH1-like (ASH1L) are expressed in hematopoietic stem and progenitor cells (HSPC), regulate HOX gene expression, and support the proliferation and/or survival of hematopoietic stem cells (4–7). ASH1L genetically cooperates with MLL to maintain hematopoiesis, as their compound deficiency results in more severe hematopoietic failure than knockout of either alone (7). At a biochemical level, MLL and ASH1L methylate lysine 4 and lysine 36, respectively, of histone H3 (8, 9); however, the underlying molecular mechanisms for their synthetic genetic interactions are unknown.

Misregulation of histone methylation caused by perturbations of epigenetic writers, readers, or erasers results in aberrant gene expression that can lead to the induction or progression of human cancers. Notably, the MLL gene is a frequent target of chromosomal translocations in acute leukemias (10). In leukemogenic translocations, MLL loses its methyltransferase activity and acquires novel epigenetic functionalities due to fusions with various partner proteins (10–13). Furthermore, MLL oncogene–mediated leukemia retains a dependence on several MLL binding partners, including lens epithelium-derived growth factor (LEDGF; ref. 14), a chromatin-associated protein implicated in cancer, autoimmunity, and HIV pathogenesis (15). LEDGF serves an essential role in MLL-dependent transcription and leukemic transformation (14) mediated in part through a conserved PWWP domain that is also present in a variety of chromatin-associated proteins and recently reported to bind methylated histone H3 lysine 36 (H3K36; refs. 16–20) with a focus on H3K36 trimethylation (H3K36me3), suggesting that LEDGF is a histone code reader. H3K36 methylation is associated with transcriptional activation and elongation, underscored by a progressive shift from H3K36 dimethylation at promoters to trimethylation at the 3′ ends of actively transcribed genes (21). Although associated with actively expressed genes, H3K36me3 has been reported to recruit histone deacetylase RPD3S and transcription corepressor ZMYND11 in yeast and human, respectively, to maintain a repressive chromatin environment during transcriptional elongation (22, 23). However, the underlying mechanism for how H3K36 dimethylation may directly contribute to transcription activation is largely unknown.

Here, we show that ASH1L maintains the histone H3K36 dimethylation (H3K36me2) mark, which is preferentially read by LEDGF to recruit/stabilize MLL on promoter proximal chromatin to activate crucial target gene expression in leukemia cells. All of these molecular events, and leukemogenesis, are specifically antagonized by the histone demethylase KDM2A. Furthermore, H3K36me3 is not required for LEDGF and MLL occupancy; rather, its reduction actually enhances recruitment of LEDGF and MLL to chromatin. Our studies reveal ASH1L, LEDGF, and KDM2A as crucial writers, readers, and erasers, respectively, of H3K36 dimethylation underlying MLL leukemia pathogenesis, illuminate the molecular mechanism by which H3K36 dimethylation directly contributes to transcription activation, and identify ASH1L as a novel target for molecular therapy of acute leukemia.

The histone binding properties of LEDGF were initially assessed using nuclear magnetic resonance (NMR) spectroscopy, which showed chemical-shift changes in the PWWP domain following titration of H3K36me3 and H3K36me2 peptides (Fig. 1A; Supplementary Fig. S1A). However, high dissociation constants (e.g., 4.4 mmol/L for the H3K36me3 complex; Supplementary Fig. S1B) suggested a weak interaction consistent with our inability to detect interactions in peptide pulldown assays, which are less sensitive than NMR (Supplementary Fig. S1C and S1D). Conversely, chromatin pulldown assays showed that recombinant LEDGF bound robustly to native mononucleosomes purified from HeLa cells (Fig. 1B; Supplementary Fig. S1E). Nucleosome interaction was abolished by mutations (W21A or F44A) of conserved aromatic residues in the hydrophobic cavity of the PWWP domain (Fig. 1B). In pulldown assays using nucleosomes purified from yeast strains containing different histone lysine mutations (Supplementary Table S1), nucleosomes containing a K36A mutation of histone H3 were not bound by GST-LEDGF, indicating that H3K36 is required for the interaction of LEDGF with nucleosomes (Fig. 1C; Supplementary Fig. S1E).

A protein–protein chromatin immunoprecipitation (ChIP) assay using THP-1 leukemia cells showed that wild-type (WT) LEDGF associated strongly with nucleosomes containing H3K36me2 in the context of physiologic chromatin, but only weakly with H3K36me1 or H3K36me3, and not at all with other methylated histones, suggesting that H3K36me2 is the highly preferred in vivo target of LEDGF (Fig. 1D; Supplementary Fig. S1F). The H3K36me2 interaction was mostly abolished by the W21A mutation that disrupted nucleosome binding in vitro (Fig. 1D). Thus, LEDGF preferentially binds dimethylated histone H3K36 in the context of nucleosomes in vivo.

ChIP coupled with high-throughput sequencing (ChIP-seq) was performed to determine the genome-wide occupancy of LEDGF, and the extent of its overlap with H3K36me2 and MLL in the MV4-11 human leukemia cell line with MLL–AF4 translocation. Approximately 89% of LEDGF-bound genes genome-wide overlapped with MLL and H3K36me2 co-occupied genes. High LEDGF and H3K36me2 occupancies were observed in genes enriched with MLL (Fig. 2A). The average distribution of all LEDGF-bound peaks recapitulated the genomic distribution of MLL and H3K36me2 (Fig. 2B) peaking around the transcription start site (TSS), in contrast to H3K36me3 distribution, which peaked at the transcription termination site (TTS). High MLL occupancies were observed in genes enriched with LEDGF, whereas MLL occupancies were much lower in genes lacking LEDGF enrichment (Fig. 2C). Comparison of ChIP-seq peaks on selected MLL target genes revealed nearly complete overlap in regional occupancy for MLL, LEDGF, and H3K36me2 around TSSs (Fig. 2D), which was further confirmed by ChIP-qPCR analyses (Supplementary Fig. S2A). These results indicate that LEDGF colocalizes with MLL and H3K36me2 around the TSSs of MLL target genes.

The observed co-occupancies raised the possibility that LEDGF functions as a reader of the histone code to recruit and/or stabilize MLL at H3K36me2-enriched promoter proximal chromatin. To test this, ChIP assays were performed in MLL–AF9-transformed HSPCs transduced with LEDGF or control shRNAs. LEDGF was downregulated to 5% to 10% of control levels (Fig. 2E) without affecting WT MLL and MLL–AF9 expression (Supplementary Fig. S2B). Using an antibody that specifically recognizes WT MLL, but not fusion MLL, ChIP assays demonstrated significantly reduced binding of WT MLL at promoter proximal regions of MLL target genes, and throughout the Hoxa9 locus, upon LEDGF knockdown (Fig. 2F; Supplementary Fig. S2C), concordant with substantially reduced transcript levels (Fig. 2E). Unexpectedly, occupancy of the MLL fusion protein (MLL–AF9) was consistently increased at the respective target loci in ChIP assays using either an anti-AF9 antibody or an anti-FLAG antibody to detect MLL–AF9 or FLAG-tagged MLL–AF9, respectively (Fig. 2F; Supplementary Fig. S2C and S2D). Furthermore, a missense mutant of MLL–AF9 (F129A) that cannot interact with LEDGF (14) retained an ability to associate with MLL target genes (Supplementary Fig. S2E). Decreased occupancies of WT MLL and increased occupancies of MLL fusion proteins were also observed for MLL–AF10 and MLL–ENL in LEDGF knockdown cells (Supplementary Fig. S2D). These results indicate that LEDGF is required for retention of WT MLL, but not MLL fusion proteins, at target gene loci in MLL-transformed HSPCs. Consistent with these results, MLL oncogene–mediated leukemogenesis is critically dependent on the WT MLL allele (24).

Despite reduced occupancy of WT MLL at target gene loci following LEDGF knockdown, H3K4me3 levels were not altered (Supplementary Fig. S2F). This is consistent with previous studies showing that knockout of MLL in HSPCs has no effect on H3K4 methylation at target genes, and the histone methyltransferase activity of MLL is dispensable for leukemogenesis (25). Rather, MLL regulates target gene expression by recruitment of the acetyltransferase MOF, which forms a stable complex with WT MLL but not MLL fusion proteins and acetylates chromatin at histone H4 lysine 16 (H4K16ac) to recruit the BRD4/pTEFb complex and facilitate transcriptional elongation (25–27). Notably, histone H4K16ac levels were reduced at Hoxa9 and Meis1 loci in LEDGF knockdown cells, and the chromatin occupancies of BRD4, pTEFb, and elongating POL II (serine 2 phosphorylated) were substantially decreased (Fig. 2F).

The foregoing results raised questions regarding how LEDGF may affect MLL fusion protein functions that are essential for misregulation of MLL target genes and MLL-induced transformation. Several translocation partners of MLL, including AF9, coexist in higher-order protein complexes (e.g., AEP or SEC), which contain known transcriptional elongation factors such as AF4 and pTEFb (28, 29). MLL oncoproteins fused with AEP components constitutively form MLL/AEP hybrid complexes to cause sustained target gene expression, which leads to transformation of HSPCs. To investigate whether LEDGF plays a role in the formation of MLL/AEP complexes on chromatin, ChIP assays were performed for AEP components AF4 and CDK9 at the Hoxa9 and Meis1 loci in MLL–AF9-transformed HSPCs (Fig. 2F). AF4 and CDK9 occupancies were significantly reduced upon LEDGF knockdown, suggesting that recruitment of the components of MLL fusion–AEP complexes at target genes is dependent on LEDGF, although LEDGF is not required for retention of MLL fusion proteins on chromatin (Fig. 2F).

The direct interaction and genome-wide co-occupancy of MLL and LEDGF raised the possibility that LEDGF chromatin binding is MLL dependent. To test this, occupancy of LEDGF was investigated by ChIP assays in MLL–AF9-transformed HSPCs transduced with MLL or control shRNAs. Knockdown of MLL (Fig. 2G) resulted in reduced MLL and LEDGF binding at promoter proximal regions of MLL target genes (Fig. 2H), indicating that MLL promotes and/or stabilizes LEDGF chromatin binding. Thus, LEDGF and MLL target association is mutually dependent, which may account for their promoter proximal co-occupancies.

Given the oncogenic requirement for LEDGF (14) and its ability to selectively bind H3K36me2 in the context of nucleosomes, we assessed the potential role for H3K36 dimethylation in MLL-transformed cells. Several histone methyltransferases mediate H3K36 dimethylation (21), most notably ASH1L, which genetically cooperates with MLL to maintain hematopoiesis (7).

ChIP-seq of MV4-11 leukemia cells showed that ASH1L displayed overlapping distribution profiles with MLL, LEDGF, and H3K36me2 around TSSs at MLL target genes (Fig. 3A and B; Supplementary Fig. S3A and B), which was further confirmed by ChIP-qPCR analyses of selected MLL target genes (Supplementary Fig. S3C). Gene set enrichment analysis (GSEA) showed that the subset of genes co-occupied by MLL, LEDGF, and ASH1L were highly enriched in genes with elevated mRNA expression in human MLL-rearranged acute lymphoblastic leukemia (MLLr ALL) versus non-MLLr ALL (Fig. 3C). Thus, ASH1L colocalizes with H3K36me2, MLL, and LEDGF around the TSSs of MLL target genes whose expression highly distinguishes MLLr human leukemias.

Knockdown of ASH1L in MLL-transformed HSPCs (Fig. 3D; Supplementary Fig. S3D) resulted in reduced H3K36me2 but not H3K36me3 levels at MLL target gene promoters (Fig. 3E; Supplementary Fig. S3E), which correlated with reduced LEDGF binding (Fig. 3F) and Hoxa9 transcript levels (Fig. 3D). ASH1L knockdown had no effect on global levels of H3K36me2 and H3K36me3 (Supplementary Fig. S3F). Occupancy of WT MLL, but not fusion MLL, was reduced at key target gene promoters and throughout the Hoxa9 locus (Fig. 3F; Supplementary Fig. S3G), and ASH1L was required for AEP complex recruitment, maintenance of H4K16ac levels, and chromatin occupancies of BRD4 and elongating POL II (Fig. 3F). These results phenocopied those observed for LEDGF knockdown, consistent with its specific binding to nucleosomes marked by H3K36me2. Thus, ASH1L maintains histone H3K36 dimethylation and promotes LEDGF and MLL recruitment at key target genes.

Activating mutations in NSD2, an H3K36 dimethyltransferase, were found in several pediatric ALL subtypes (30, 31), raising the question of whether NSD2 plays any role in LEDGF and MLL recruitment. Efficient knockdown of NSD2 in MLL–AF10-transformed HSPCs resulted in modest reductions of LEDGF and MLL occupancy at the Hoxa9 gene promoter, but not at the Meis1 gene promoter (Supplementary Fig. S3H), which correlated with modest reduction of Hoxa9, but not of Meis1, transcript levels (Supplementary Fig. S3I). This contrasted with ASH1L knockdown, which caused major, consistent reduction of LEDGF and MLL recruitment at MLL target genes.

The requirement for ASH1L and H3K36 dimethylation in MLL-mediated oncogenesis was initially assessed in MLL-transformed HSPCs. Cells transduced with shRNAs targeting ASH1L (Fig. 4A) showed morphologic and phenotypic differentiation and underwent apoptosis (Fig. 4B–D; Supplementary Fig. S4A and S4B). ASH1L knockdown substantially compromised the in vitro growth and clonogenic activity of HSPCs transformed by MLL oncogenes, but did not affect cells transformed by a non-MLL oncogene (Fig. 4E). In transplantation assays, MLL–AF10-transformed HSPCs knocked down for ASH1L showed a substantial impairment in their ability to induce leukemia, as evidenced by a marked increase in disease latency and reduced penetrance (Fig. 4F).

The role of ASH1L in MLL-mediated leukemogenesis was further investigated in human leukemia cells. Knockdown of ASH1L resulted in reduced HOXA9 expression in MLLr leukemia cell lines (Fig. 5A), which correlated with reduced growth in methylcellulose and liquid cultures, whereas non-MLLr leukemia cells were unaffected (Fig. 5B and C). Xenotransplantation studies showed that knockdown of ASH1L markedly impaired bone marrow engraftment efficiency of MV4-11 cells in NSG mice by over 90% compared to control (shLuc) cells at 4 weeks (Fig. 5D). Similarly, xenotransplantation studies of primary MLLr acute myeloid leukemias (AML) showed that knockdown of ASH1L substantially reduced engraftment efficiency (∼80%) compared to control at 10 weeks after transplant (Fig. 5E). Thus, ASH1L is required for MLL oncogene–induced leukemogenesis.

In protein–protein ChIP assays, LEDGF associated preferentially with H3K36me2, which is essential for MLL-mediated oncogenesis, and weakly with H3K36me3 (Fig. 1D). The potential contribution of H3K36 trimethylation to MLL-mediated transformation was assessed by knocking down SETD2, the only known histone H3K36me3 methyltransferase (21). SETD2 knockdown in MLL–AF9-transformed HSPCs (Fig. 6A) reduced global levels of H3K36me3 (Fig. 6B). Compared to ASH1L knockdown, SETD2 knockdown had minor effects on Hoxa9 gene expression (Fig. 6A) and clonogenic activity of MLL-transformed HSPCs (Fig. 6C). SETD2 plays an important role in maintaining genome stability (32), which may account for partially reduced leukemia cell growth.

ChIP assays showed that SETD2 knockdown resulted in reduction of H3K36me3 levels at MLL target genes (Fig. 6D). In contrast, H3K36me2 was increased, which correlated with increased LEDGF and MLL occupancy, further supporting the key role of H3K36me2 in MLL/LEDGF recruitment. Fusion MLL was maintained at key target gene promoters in SETD2 knockdown cells (Fig. 6D). Thus, SETD2 and H3K36 trimethylation do not directly regulate MLL oncogene–induced leukemogenesis.

The role of reduced H3K36me2 through forced demethylation was also assessed in MLL-mediated transformation. Several families of histone demethylases demethylate H3K36me2 in mammalian cells (1). Among them, the KDM2 family is of particular interest because genetic evidence indicates antagonism between ASH1 and dKDM2 in Drosophila (33). In humans, two paralogs (KDM2A and KDM2B) preferentially demethylate H3K36me2 and repress transcription (1, 34).

Overexpression of KDM2A, but not KDM2B, in MLL–AF10-transformed HSPCs resulted in reduced Hoxa9 expression (Fig. 7A; Supplementary Fig. S5A). KDM2A overexpression also resulted in reduced H3K36me2 but not H3K36me3 levels at MLL target genes and reduced occupancy of LEDGF and WT MLL, but not fusion MLL (Fig. 7B). ASH1L binding to MLL target genes was not significantly affected by forced KDM2A expression (Supplementary Fig. S5B). KDM2A overexpression had no effect on global levels of H3K36me2 and H3K36me3 (Supplementary Fig. S5C). Thus, KDM2A demethylates H3K36me2 and promotes the chromatin dissociation of MLL and LEDGF, concomitant with reduced expression of MLL target genes.

Overexpression of KDM2A, but not KDM2B, induced monocytic differentiation of MLL–AF10-transformed cells (Supplementary Figs. S5D) and markedly reduced colony-forming cell activity of MLL-transformed cells, whereas cells transformed by E2A–HLF, which acts through non-HOX pathways, were unaffected (Fig. 7C and D). Forced expression of KDM2A also antagonized the leukemogenic properties of MLL–AF10-transformed HSPCs, causing increased latency and reduced penetrance in transplant assays (Fig. 7E). These results demonstrate that KDM2A specifically opposes MLL-induced leukemia, further supporting the role for H3K36me2 in MLL-mediated oncogenesis.

The epigenetic pathways and mechanisms involved in MLL leukemia pathogenesis are multiple and complex, and the effectiveness of early-stage targeted molecular therapies is not yet established. The recent discovery of BET inhibitor resistance further underscores the need for discovery of new drug targets and development of combined therapies (35, 36). Our studies are the first to define the epigenetic interactions involving placement, interpretation, and removal of H3K36me2 underlying MLL leukemia pathogenesis. Furthermore, our studies are particularly notable for implicating ASH1L as a promising target for novel molecular therapy in addition to providing important mechanistic insight into the role of H3K36 dimethylation in transcription.

Methyl lysine residues “written” by histone methyltransferases are recognized and bound by “reader” proteins that recruit various components of the transcription complex to translate the histone code into specific gene expression states (2). In yeast, H3K36me3 recruits histone deacetylase RPD3S to reset the coding region of actively transcribed genes to a relatively repressive state, thus suppressing intragenic transcription initiation (23). In human, H3K36me3 recruits the transcription corepressor ZMYND11 to highly expressed genes to fine-tune transcription by modulating RNA polymerase II at the elongation stage (22). It remains unknown how H3K36me2 may directly contribute to activation. Our studies answer this long-standing question and support a model in which ASH1L writes the H3K36me2 mark, which is read by LEDGF to recruit/stabilize MLL on chromatin to maintain crucial target gene expression in leukemia cells. These findings delineate the functional interrelationships of two lysine histone methyltransferases involved in MLL leukemia pathogenesis and illuminate their genetic interactions as evolutionarily conserved trithorax group proteins. Together they counteract Polycomb group mediated silencing to regulate cell fate through maintaining homeotic gene expression (3). This provides a molecular context for the conserved interactions of trxG proteins in development, hematopoiesis, and leukemogenesis.

The role of LEDGF as a histone code reader that mediates chromatin association of MLL is consistent with previous studies showing that its contribution in MLL leukemia requires its conserved PWWP domain (14, 17). The latter has recently been reported to bind H3K36me2/3 (16–19) with a focus on H3K36me3; however, our data strongly support H3K36me2 as the biologically relevant target. The differences of our results with previous studies (18, 37) are likely accounted for by prior use of ChIP conditions not efficiently precipitating the MLL-associated form of LEDGF (18) and more limited genomic assessment using customized arrays covering only Hox loci and several other developmental genes (18, 37). Although both the tri- and di- H3K36 methylation histone marks have been implicated in transcription activation (21), LEDGF preferentially binds H3K36me2 in the context of physiologic chromatin in human leukemia cells. In support of this, LEDGF and MLL colocalize with H3K36me2 peaking at the TSS, in contrast to H3K36me3 distribution, which peaks at the TTS. Furthermore, knockdown of SETD2, the only known H3K36 trimethyltransferase, reduced H3K36me3 levels genome-wide and particularly at MLL target genes, whereas H3K36me2 was respectively increased, which correlated with increased LEDGF and MLL occupancy. Consistent with the lack of a role for SETD2 and H3K36me3 in MLL-dependent transcription, SETD2-inactivating mutations occur with a significantly higher frequency in patients with MLL-rearranged leukemia and contribute to leukemogenesis through a global reduction of H3K36me3 (38, 39). Our results indicate that a molecular consequence of prevalent SETD2 loss-of-function mutations in MLL leukemias is increased occupancy of H3K36me2, LEDGF, and MLL at MLL target genes. Thus, H3K36me3 is not essential for MLL occupancy and leukemogenesis; rather, its relative reduction actually enhances molecular features that sustain MLL leukemia.

An unexpected observation is that ASH1L and LEDGF are not required for the chromatin association of MLL fusion proteins at crucial leukemia target genes. In fact, depletion of either ASH1L or LEDGF resulted in increased occupancy of MLL fusion proteins, suggesting a likely competition for chromatin sites vacated by WT MLL. Both WT MLL and MLL fusion proteins are required for sustained expression of genes that are essential for leukemogenesis (24); however, it is not fully understood how they are recruited to chromatin. The differential dependence on LEDGF for occupancy at target gene loci in MLL-transformed cells indicates that WT and fusion MLL proteins have different recruitment mechanisms. In support of this, the recruitment of WT MLL requires both its CXXC/RD1 domain and PHD3 finger, which mediate direct interactions with the PAF1 elongation complex and H3K4me3, respectively. However, the CXXC/RD1 domain alone, which binds CpG-rich DNA as well as PAF1, is sufficient for the recruitment of MLL fusion proteins (40). Thus, WT MLL has more complex requirements for chromatin association, including a selective dependence on LEDGF and H3K36me2, which appear to be bypassed by MLL fusion proteins. Although LEDGF is not required for the chromatin occupancy of MLL fusion proteins, their leukemogenic activity is critically dependent on LEDGF, with which they specifically interact (14). This raised a question regarding whether LEDGF has a recruitment-independent role in modulating MLL fusion protein functions in their roles as components of aberrant AEP/SEC complexes, which contain transcription elongation factors, including MLL fusion partners, essential for leukemia. Our data show that the chromatin association of AEP/SEC components AF4 and CDK9 is significantly reduced upon LEDGF knockdown, suggesting that the recruitment of components of the fusion protein complex at target genes is dependent on LEDGF, although LEDGF is not required for MLL fusion protein retention on chromatin.

The dependence on ASH1L establishes it as a candidate target for molecular therapy of MLLr acute leukemias, which are generally associated with a poor prognosis (10). Our results show that ASH1L is particularly enriched at a subset of genes (e.g., HOXA9, MEIS1, and CDK6) that are differentially expressed in MLLr leukemias and crucial for leukemia pathogenesis. Their constitutive expression is mediated by the combined actions of MLL WT and fusion proteins (24), and targeting either factor effectively antagonizes MLL leukemia. Although small-molecule inhibitors are not yet available, genetic studies suggest that ASH1L inhibition may not be unmanageably toxic. Homozygous ASH1L mutation was reported to result in decreased long-term hematopoietic stem cell numbers; however, increased self-renewal of progenitors compensated for HSC loss and sustained relatively normal mature hematopoietic cell output (7). Partial reduction in ASH1L activity shows greater cytotoxicity for MLL leukemia cells, defining it as a selective target for therapeutic intervention of leukemia. Future studies are warranted when inhibitors are developed to further assess the efficacy of targeting ASH1L as a therapeutic strategy in MLLr leukemia and possibly other cancer types dependent on elevated HOX gene expression.

Maintenance of HOX gene expression and MLL oncogene–induced leukemogenesis is opposed by the histone code “eraser” KDM2A, a demethylase that counteracts the actions of ASH1L. This parallels results in Drosophila, where dKDM2 is a component of the dRING-associated factor complex, a Polycomb group silencing complex, and cooperates with Polycomb to counteract homeotic gene activation by trxG histone methyltransferases TRX and ASH1 (33). In humans, KDM2 has two homologs (KDM2A and KDM2B) that demethylate H3K36me2 and repress transcription (41, 42). KDM2A interacts with SUZ12, a component of Polycomb repressive complex 2 (43). Overexpression of KDM2A reduced MLL-dependent transcription and leukemic transformation. KDM2A demethylates H3K36me2 at MLL target genes, and promotes the chromatin dissociation of MLL and LEDGF, elucidating a molecular pathway for how KDM2A counteracts trxG proteins to repress transcription.

The action of KDM2A in suppressing MLL leukemia by opposing ASH1L activity may reflect an analogous role in normal hematopoiesis. KDM2A transcripts are low in HSPCs and increase with myeloid differentiation, which is the inverse of expression profiles for MLL, LEDGF, and ASH1L (Microarray Database of Gene Expression Commons). KDM2A is recruited to nonmethylated CpG islands through its CXXC domain, where it removes histone H3K36me2 (41, 44). Our studies show that ASH1L, MLL, and LEDGF co-occupy CpG islands at MLL target genes in leukemia cells. Taken together, these findings support a model in which KDM2A initiates differentiation of normal and leukemic myeloid progenitors by binding to nonmethylated CpG islands, resulting in removal of histone H3K36me2, which in turn promotes the dissociation of MLL and LEDGF, and subsequent silencing of MLL target genes.

Phoenix and 293T cells were cultured in Dulbecco's Modified Eagle Medium (DMEM) supplemented with 10% FBS and penicillin–streptomycin–glutamine. Human leukemia cell lines were cultured in RPMI-1640 medium supplemented with 10% FBS and penicillin–streptomycin–glutamine. Cell lines were obtained from the ATCC or DSMZ in 2013 and not further authenticated. Cell lines were tested periodically for Mycoplasma by PCR (MD Bioproducts), and results were verified by IDEXX RADIL Laboratories.

ChIP assays using mouse myeloid progenitors transformed by MLL fusion genes were performed using 0.5 × 10⁷ to 1 × 10⁷ cells per IP under conditions described previously (45). ChIP-qPCR signals were calculated as a percentage of input. Primer sequences and antibodies used in ChIP assays are provided in Supplementary Tables S2 to S4.

Whole-genome ChIP using MV4-11 cells was performed under conditions described previously (45) using 2 × 10⁷ to 4 × 10⁷ cells per IP with specific antibodies (Supplementary Table S4). ChIP-seq libraries were prepared according to the Illumina protocol and sequenced using Illumina HiSeq. Sequencing reads were mapped to the human reference genome (hg19) using Bowtie with parameters: -v 2 -k 2 -m 1 –best –strata. Clonal reads were removed before enriched peak/region calling with MACS version 2.1.0. Enriched regions (fold change, ≥ 5) were filtered with false discovery rate ≤ 0.01 to be considered as statistically significant ChIP peaks. Downstream analyses including genome-wide localization density and average profile plots were performed by using ngs.plot and R/Bioconductor.

MLL–AF10-transformed mouse HSPCs were transduced with control or appropriate shRNAs against LEDGF or ASH1L. Following puromycin selection, cells (1 × 10⁶) were transplanted intravenously into lethally irradiated C57BL/6 mice (900 rads) with 2 × 10⁵ syngeneic bone marrow cells. Similar transplant conditions were used for leukemogenesis assays using MLL–AF10-transformed mouse HSPCs transduced with KDM2A expression vector. All experiments on mice in this study were performed with the approval of and in accordance with Stanford University's Administrative Panel on Laboratory Animal Care.

Xenotransplant assays were performed as previously described (46). MV4-11 human leukemia cells were transduced with control or ASH1L shRNA virus by spinoculation. After 48 hours, 4 million cells in 0.2 mL PBS containing 0.5% FBS were transplanted into sublethally irradiated (200 rad) NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NSG) mice (8–10-week females). After 4 weeks, the bone marrow was collected and analyzed for the presence of transduced (GFP⁺) human leukemia cells by flow cytometry.

Human AML samples were obtained from patients at the Stanford University Medical Center (Stanford, CA) with informed consent and institutional review board approval in accordance with the Declaration of Helsinki. Primary AML specimens (MLLr) were thawed and cultured overnight in myelocult medium containing human cytokines (100 ng/mL SCF, 100 ng/mL FLT3, 50 ng/mL TPO, 20 ng/mL IL3, 20 ng/mL IL6, and 1 μmol/L SR1). In triplicate, three million cells were transduced with ASH1L shRNA virus using retronectin (Clontech) according to the manufacturer's instructions. After 48 hours, 2 million cells in 0.2 mL PBS containing 0.5% FBS were transplanted into sublethally irradiated (200 rad) NSG mice. After 10 weeks, the bone marrow was collected and analyzed for the presence of transduced (GFP⁺) human leukemia cells by flow cytometry.

The WT LEDGF PWWP (residues 1–120) was expressed in E. coli BL21(DE3) pLysS grown in ¹⁵NH₄Cl-supplemented M9-minimal medium. Bacteria were harvested by centrifugation after IPTG induction and lysed via sonication. The GST fusion protein was purified on a glutathione Sepharose 4B column, cleaved with PreScission protease, and concentrated in Millipore concentrators using 25 mmol/L Tris pH 6.8 buffer containing 150 mmol/L NaCl and 8% ²H₂O.

NMR experiments were performed at 298 K on a Varian INOVA 600-MHz spectrometer using uniformly ¹⁵N-labeled LEDGF PWWP. The histone binding was characterized by monitoring chemical shift changes in ¹H,¹⁵N heteronuclear single quantum coherence (HSQC) spectra of 0.1 to 0.2 mmol/L PWWP as histone peptides were titrated in. The spectra were processed with NMRPipe and analyzed with nmrDraw. The dissociation constant (Kd) was determined by a nonlinear least-squares analysis using a Kaleidagraph program and the equation: Δδ = Δδmax((([L] + [P] + KD) − sqrt(sqr([L] + [P] + KD) − (4*[P]*[L])))/(2*[P])), where [L] is concentration of the peptide, [P] is concentration of PWWP, Δδ is observed chemical shift change, and Δδmax is the difference in chemical shifts of the free and the ligand-bound protein.

Yeast nucleosomes were purified as described previously (47). At the last step, nuclei were resuspended in 400 μL NP buffer (20 mmol/L Tris-HCL pH 8.0, 150 mmol/L NaCl, 5 mmol/L KCl, 1 mmol/L EDTA, 5 mmol/L CaCl₂, 10% glycerol, 1 mmol/L PMSF, protease inhibitors). Micrococcal nuclease was added to a final concentration of 10 u/mL, and incubated for 5 minutes at 37°C to obtain nucleosomes. The reaction was stopped with 10 μL of 0.5 mol/L EDTA. After spinning with maximum speed for 20 minutes at 4°C, the supernatant that contained nucleosomes was aliquoted and kept at −80°C. Yeast strains are listed in Supplementary Table S1.

Histone peptide binding assays were performed as described (48). For mononucleosome binding assays, 1 μg of mononucleosome purified from HeLa cells or S. cerevisiae and 5 μg of purified GST-LEDGF WT or mutant proteins were incubated in 300 μL binding buffer (300 mmol/L NaCl, 0.1% NP40, and 50 mmol/L Tris pH 7.5, 10% glycerol) at 4°C for 4 hours, followed by an additional 1 hour with glutathione beads (Amersham). After extensive washing with binding buffer, beads were boiled with SDS buffer and subjected to Western blot analysis using anti-H3 antibody (Abcam; ab1791).

Modified protein–protein ChIP assays using THP-1 human leukemia cells for detection of in situ LEDGF–histone interactions were performed as described previously (48).

Raw microarray data of MLL-rearranged (20 cases) and MLL germline ALL (112 cases) (49) were processed and normalized by the Robust Multi-Array (RMA) method using R/Bioconductor package “affy.” Normalized expression values were subjected to GSEA as described previously (50).

pMSCV-neo constructs encoding MLL–AF9, MLL–AF10, MLL–ENL, and E2A–HLF were described previously (14). Expression vectors for FLAG-KDM2A and FLAG-KDM2B were constructed by PCR using pCMV-SPORT6-KDM2A and pCMV-SPORT6-KDM2B (Open Biosystems) as templates and cloned into pMSCV-puro vector (Clontech).

ASH1L (human) shRNA constructs were purchased from Open Biosystems (Supplementary Table S5). For knockdown of LEDGF, ASH1L, and SETD2 in mouse cells, shRNAs (Supplementary Fig. Table S5) were cloned into the lentiviral pLKO.1 vector. Lentivirus was produced by cotransfection of viral vectors with pCMV–dR8.2 and pCMV-VSVG packaging constructs into 293T cells. Medium containing virus was collected 48 hours after transfection and used for transductions.

Total RNA was isolated and converted to cDNA as described previously (14), then subjected to RT-PCR on a CFX384 real-time PCR system (Bio-Rad) using TaqMan assays. Relative expression was normalized to β-Actin or Gapdh. TaqMan probes used in real-time PCR are provided in Supplementary Table S6.

HSPC transformation assays were performed as previously described (14). For knockdown experiments, MLL fusion–transformed cells were transduced with shRNA lentivirus by spinoculation, cultured in methylcellulose medium for 2 days, and then selected in 2 μg/mL puromycin for 3 days, before harvesting for further analysis.

O. Gozani has ownership interest (including patents) in EpiCypher, Inc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L. Zhu, M.L. Cleary

Development of methodology: L. Zhu, J. Shen, D. Schneidawind, R. Majeti

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Zhu, B.J. Klein, M. Onishi, D. Schneidawind, C. Buechele, J. Duque-Afonso, F. Zhu, G. Mas Martin, T.G. Kutateladze, M.L. Cleary

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Zhu, Q. Li, S.H.K. Wong, M. Huang, B.J. Klein, C. Buechele, L. Hansen, J. Duque-Afonso, F. Zhu, T.G. Kutateladze, M.L. Cleary

Writing, review, and/or revision of the manuscript: L. Zhu, S.H.K. Wong, B.J. Klein, D. Schneidawind, C. Buechele, J. Duque-Afonso, M.L. Cleary

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Zhu, S.H.K. Wong, L. Ikenouye, M. Onishi, D. Schneidawind, M.L. Cleary

Study supervision: L. Zhu, M.L. Cleary

Other (provided reagents and supervised experiments related to those reagents): O. Gozani

The authors thank Cita Nicolas and Maria Ambus for technical support, Ziming Weng and Dr. Eliezer Calo for support on ChIP sequencing, Dr. Xianxin Hua for MLL knockdown constructs, and Dr. Matt Barber for suggestions on nucleosome binding assays. They thank members of the Cleary lab for discussions. They also acknowledge the Stanford Hematology Division Tissue Bank and the patients for providing clinical samples.

The Gene Expression Omnibus accession number for the ChIP-seq data reported in this article is GSE73528.

This work was supported by NIH grants GM106416 (T.G. Kutateladze), GM079641 (O. Gozani), and CA116606 (M.L. Cleary). J. Duque-Afonso was supported by the German Research Foundation (Deutsche Forschungsgemeinschaft, ref. DU 1287/2-1). C. Buechele and D. Schneidawind were supported by the Dr. Mildred Scheel Stiftung. The authors also acknowledge support from the Children's Health Initiative of the Packard Foundation, the California Institute for Regenerative Medicine, and the Child Health Research Institute.